Systematic design of
advanced control structures
Adriana Reyes-Lúa
February 28th, 2020
No
rwe
gia
n U
niv
ers
ity
ofS
cie
nce
an
d T
ech
no
log
y
2
Content
• Motivation and scope
• Active constraint switching with advanced control structures (chapter 2)
– Case study: mixing
– Case study: distillation column
– Case study: cooling cycle (chapter 3)
– Case study: cooler (chapter 4)
• MV to MV constraint switching
– Split range control
• Design of standard split range controllers (chapter 5)
• Generalized split range controller (chapter 6)
– Multiple controllers with different setpoints (chapter 7)
• Improved PI control for tank level (chapter 8)
• Conclusions
3
Motivation and scope
4
Motivation and scope
CV: controlled
variable (output, y)• Temperature
• Pressure
• Concentration
5
Motivation and scope
CV: controlled
variable (output, y)• Temperature
• Pressure
• Concentration
MV: manipulated
variable (input, u)• Valve opening
• Compressor rotational
speed
6
Motivation and scope
CV: controlled
variable (output, y)• Temperature
• Pressure
• Concentration
MV: manipulated
variable (input, u)• Valve opening
• Compressor rotational
speed
DV: disturbance variable (d)• Ambient temperature
• Raw materials
• Desired production
7
Motivation and scope
CV: controlled
variable (output, y)• Temperature
• Pressure
• Concentration
MV: manipulated
variable (input, u)• Valve opening
• Compressor rotational
speed
DV: disturbance variable (d)• Ambient temperature
• Raw materials
• Desired production
8
Motivation and scope
9
Motivation and scope
Top-down analysis:S1-S4: Identify steady-state
optimal operation
Bottom-up analysis:S5-S7: Design control
structure
10
Motivation and scope
Bottom-up analysis:S5: regulatory control layer
11
Motivation and scope
Bottom-up analysis:S5: regulatory control layer
S6: supervisory control layer
S7: online optimization layer
12
Motivation and scope
Bottom-up analysis:S5: regulatory control layer
S6: supervisory control layer
S7: online optimization layer
13
Motivation and scope
Bottom-up analysis:S5: regulatory control layer
S6: supervisory control layer
S7: online optimization layer
Keeps operation
in the right
active constraint region
14
Motivation and scope
Constraint region«region in the disturbance
space defined by which
constraints are active within it»
Disturbance 1
Dis
turb
an
ce
2
Jacobsen and Skogestad (2011) Active constraint regions for optimal operation of chemical processes. Industrial & Engineering Chemistry Research.
S6: supervisory control layer
15
Motivation and scopeS6: supervisory control layer
16
Motivation and scope
Model predictive control
S6: supervisory control layer
Cost
Model
Constraints
17
Motivation and scope
Model predictive control
Advanced control structures
S6: supervisory control layer
18
Active constraint switching with classical advanced
control structures
Figure taken from www.transmittershop.com/blog/causes-solutions-annoying-noise-control-valves
19
Active constraint switching with classical advanced
control structures
Figure taken from www.transmittershop.com/blog/causes-solutions-annoying-noise-control-valves
20
Active constraint switching with classical advanced
control structures
Figure taken from www.transmittershop.com/blog/causes-solutions-annoying-noise-control-valves
21
Design procedure for active constraint switching with
classical advanced control structures
A1• Define control objectives, CV constraints and MV constraints
A2• Organize constraints in priority list
A3• Identify possible and relevant active constraint switches
A4• Design control structure for optimal operation
A5• Design control structure to handle active constraint switches
22
Design procedure for active constraint switching
Case study:Mixing of
air and MeOH
24
Design procedure for active constraint switching
Step A1: Define control objectives, CV constraints and MV constraints
25
Design procedure for active constraint switching
Step A1: Define control objectives, CV constraints and MV constraints
2 CVs2 MVs
26
Design procedure for active constraint switching
Step A1: Define control objectives, CV constraints and MV constraints
Control objectives:• Keep y1 = xMeOH = 0.10 ideal
• Keep y1 =xMeOH > 0.08
• Control y2 = mtot ideal
2 CVs2 MVs
27
Design procedure for active constraint switching
Control objectives:• Keep y1 = xMeOH = 0.10
• Keep y1 =xMeOH > 0.08
• Control y2 = mtot
2 CVs2 MVs
u1 is has a maximum value
Step A1: Define control objectives, CV constraints and MV constraints
28
Design procedure for active constraint switching
Step A2: Organize constraints in priority list
Figure from www.indelac.com/blog/control-valves-vs.-regulators-in-control-applications
29
Design procedure for active constraint switching
Step A2: Organize constraints in priority list
• Constraint on air flow (u1)
• Constraint on MeOH flow (u2)(P1) Physical MV inequality constraints
• Constraint (max and min) on xMeOH (y1)(P2) Critical CV inequality constraints
• Setpoint on xMeOH (y1)(P3) Less critical CV and MV constraints
• Setpoint on mtot (y2)(P4) Desired throughput
• No unconstrained degrees of freedom(P5) Self-optimizing variables
30
Design procedure for active constraint switching
Step A3: Identify possible and relevant active constraint switches
31
Design procedure for active constraint switching
Step A3: Identify possible and relevant active constraint switches
• Case 1: CV to CV constraint switching
One MV switching between two alternative CVs.
32
Design procedure for active constraint switching
Step A3: Identify possible and relevant active constraint switches
Split range control
Valve position control
Different controllers
with different setpoints
• Case 2: MV to MV constraint switching
More than one MV for one CV.
33
Design procedure for active constraint switching
Step A3: Identify possible and relevant active constraint switches
• Case 3: MV to CV constraint switching
MV controlling a CV that may saturate; no extra MVs
Input saturation pairing rule«an MV that is likely to saturate at
steady-state should be paired with a
CV that can be given up»
Low priority CVHigh priority CV:
always controlled
MV that does not
saturate
34
Design procedure for active constraint switching
Step A3: Identify possible and relevant active constraint switches
• Case 3: MV to CV constraint switching
MV controlling a CV that may saturate; no extra MVs
Following input
saturation pairing rule
NOT following input
saturation pairing rule
35
Design procedure for active constraint switching
Step A3: Identify possible and relevant active constraint switches
• At nominal operation point all
constraints are satisfied
• Constraint switch:
• Reach maximum air flow (u1)
• Lose a degree of freedom (case 3)• Must give up controlling the
constraint with the lowest priority
(desired throughput)
36
Design procedure for active constraint switching
Step A4: Design control structure for optimal operation
37
Design procedure for active constraint switching
Step A4: Design control structure for optimal operation
Following input
saturation pairing rule
Case A
Low
priority
CV
MV likely to saturate
38
Design procedure for active constraint switching
Step A4: Design control structure for optimal operation
Case A Case B
NOT following input
saturation pairing rule
Following input
saturation pairing rule
Low
priority
CV
MV likely to saturate
High
priority CV
not always
controlled
MV likely to saturate
39
Design procedure for active constraint switching
Step A4: Design control structure for optimal operation
Case A Case B
NOT following input
saturation pairing rule
Following input
saturation pairing rule
Needs MV to CV
switching
40
Design procedure for active constraint switching
Step A5: Design control structure to handle active constraint switches
41
Design procedure for active constraint switching
Step A5: Design control structure to handle active constraint switches
Case B-SRC
Split range control+selector
Case B-VPC
Valve position control + selector
42
Design procedure for active constraint switching
Case study: Mixing of air and MeOHMV1 is saturated:
lost degree of freedomHigh priority CV: concentration
Low priority CV (throughput) MV2 is not saturated:
It should be used to control the high priority CV
43
MV to MV constraint switching
Split range control Different controllers with
different setpoints
Valve position
control
44
Classical split range control
Monogram of Instruments and Process Control
prepared at Cornell, NY, in 1945
CV
MV1
MV2
Eckman, D.P. (1945).
Principles of industrial
control, New York.
45
Classical split range control
46
Classical split range control
v internal signal to split range block limited physical meaning
v* split value
ui controller output physical meaning
αi gain from v to ui slope
47
Classical split range control
v internal signal to split range block limited physical meaning
v* split value degree of freedom
ui controller output physical meaning
αi gain from v to ui slope
48
Classical split range control
v internal signal to split range block limited physical meaning
v* split value degree of freedom
ui controller output physical meaning
αi gain from v to ui slope
50
Design of split range control: select slopes
Goal: get desired loop gain |g C|
at crossover frequency
51
Design of split range control: select slopes
Goal: get desired loop gain |g C|
at crossover frequency
52
Design of split range control: select slopes
Goal: get desired loop gain |g C|
at crossover frequency
Fast process Slow process
Desired
gain for ui
Common
gain in C
DOFDesired
gain for ui
Common
gain in C
DOF
53
Design of split range control: order of MVs
Define the desired operating point for every MV
Group the MVs according to the effect on the CV
Within each group, define order of use
55
Design of split range control
u1 = uAC : air conditioning (AC)
u2 = uCW : cooling water (CW)
u3 = uHW : heating water (HW)
u4 = uEH : electrical heating (EH)
56
Classical split range control: a compromise
1 DOF
2 tuning parameters
57
Generalized split range controller
58
Generalized split range controller
Preliminary step:
• Define order of use of MVs ( j=1,…,N)
• Tune controllers
59
Generalized split range controller
«Baton strategy» logic
k is the active input
• Ck computes uk’ (suggested value for uk)
• If ukmin< uk’< uk
max
• Keep uk active and uk uk’
• Keep remaining ui at limiting value
• else
• Set uk= ukmin or uk< uk
max, depending on the reached limit
• New active input selected according to predefined sequence
(j= k-1 or j=k+1)Preliminary step:
• Define order of use of MVs ( j=1,…,N)
• Tune controllers
60
Generalized split range controller
«Baton strategy» logic
k is the active input
• Ck computes uk’ (suggested value for uk)
• If ukmin< uk’< uk
max
• Keep uk active and uk uk’
• Keep remaining ui at limiting value
• else
• Set uk= ukmin or uk< uk
max, depending on the reached limit
• New active input selected according to predefined sequence
(j= k-1 or j=k+1)Preliminary step:
• Define order of use of MVs ( j=1,…,N)
• Tune controllers The active input will decide when to switch and
will remain active as long as it is not saturated.
61
Generalized split range controller
u1 = uAC : air conditioning (AC)
u2 = uCW : cooling water (CW)
u3 = uHW : heating water (HW)
u4 = uEH : electrical heating (EH)
62
Generalized vs standard split range controller
63
Generalized split range controller: initialization
How to
start?
k
64
Generalized split range controller: initialization
How to
start?
This suggested input was
not being applied while
input k was not in use
This accumulated error is
not due to the previous
actions of input kk
65
Generalized split range controller: initialization
Resetting:Only use error
when I receive
the baton
Initial action proportional to error at tb
k
66
Generalized split range controller: initialization
Back-calculation:
I was keeping
track of the
applied input
67
Generalized split range controller: initialization
Resetting:
Back-calculation:
68
Generalized split range controller vs MPC
69
Multiple controllers with different setpoints
Does this make sense at any point?
70
Multiple controllers with different setpoints
Setpoint
deviation
Input
usage
71
Multiple controllers with different setpoints:
Optimal setpoint deviation
Linear for u and quadratic for Δy Inputs are a linear function of output
72
Multiple controllers with different setpoints:
Optimal setpoint deviation
Linear for u and quadratic for Δy Inputs are a linear function of output
Cost when using uk as input
73
Multiple controllers with different setpoints:
Optimal setpoint deviation
Linear for u and quadratic for Δy
Optimal
setpoint deviation
minimizing cost
Inputs are a linear function of output
Cost when using uk as input
74
Multiple controllers with different setpoints: Case study
QAC : air conditioning
QHW : heating water
QEH : electrical heating
75
Multiple controllers with different setpoints: Case study
Cost: linear for u and quadratic for Δy
QAC : air conditioning
QHW : heating water
QEH : electrical heating
76
Multiple controllers with different setpoints: Case study
Cost: linear for u and quadratic for Δy
QAC : air conditioning
QHW : heating water
QEH : electrical heating
Inputs (Qi) are a linear function of output (T)
77
Multiple controllers with different setpoints: Room T
Cost: linear for u and quadratic for Δy
QAC : air conditioning
QHW : heating water
QEH : electrical heating
Inputs (Qi) are a linear function of output (T)
Optimal setpoint deviation minimizing cost
78
Multiple controllers with different setpoints: Room T
QAC : air conditioning
QHW : heating water
QEH : electrical heating
Optimal setpoint
deviation
minimizing cost
79
Multiple controllers with different setpoints: Room T
80
Multiple controllers with different setpoints: Room T
Lower accumulated
cost with minimum
setpoint deviation
Constant setpoint
81
Final comments
• Steady-state optimal operation may be easily achieved using PID-based
control structures– Chapters 2,3,4: active constraint switching
– Chapter 7: optimal setpoints
82
Final comments
• Steady-state optimal operation may be easily achieved using PID-based
control structures– Chapters 2,3,4: active constraint switching
– Chapter 7: optimal setpoints
• Useful to systematically define control objectives, feasibility and tools– Priority list of constraints
– Control structures available for each type of switch (CV-CV, MV-MV, MV-CV)
83
Final comments
• Steady-state optimal operation may be easily achieved using PID-based
control structures– Chapters 2,3,4: active constraint switching
– Chapter 7: optimal setpoints
• Useful to systematically define control objectives, feasibility and tools– Priority list of constraints
– Control structures available for each type of switch (CV-CV, MV-MV, MV-CV)
• Possible to improve performance of PID-based advanced control– Chapters 5, 6: design of split range controllers
– Chapter 8: improved level control
84
One final comment
• The “gap” between theory and practice can be in both directions
Åström, K. J., & Kumar, P. R. (2014). Control: A perspective. Automatica, 50(1), 3–43.
Centrifugal governor used in steam
engines in the 1780’s:
Proportionally controls fuel flow to
maintain engine speed.
Theoretical investigation started about
a century later.speed
fuel
Systematic design of
advanced control structures
Thank you for your attention!
No
rwe
gia
n U
niv
ers
ity
ofS
cie
nce
an
d T
ech
no
log
y